Newsletter · · Ashutosh Agarwal
Venture Splits Into a Barbell as Series A Becomes the Bloodbath - The VC Read - Week of September 3, 2026
Startups and venture for the week of September 3, 2026. Podcast synthesis on a venture market splitting into a barbell, with AI infrastructure and cheap AI-native seed rounds thriving while Series A snaps in the middle, shutdowns running at double last year's pace, investors passing on strong companies for fear of being Sherlocked by the model labs, plus the SpaceX and Cursor deal, Anthropic's first profit, Workday's take-private and a DOJ probe of a16z.
The VC Read
Week of September 3, 2026: Venture Splits Into a Barbell as Series A Becomes the Bloodbath
Venture is splitting in two. Below the app layer, money is free; above it, founders are getting "Sherlocked." And in the middle, the shutdown pile is filling up. Plus: the $60B and $7B M&A machine, Anthropic's first profit, Workday goes private, and the DOJ knocks on a16z's door.
The Big Debate: Is the middle of venture dying?
Here is the uncomfortable shape of venture right now, described by the people living inside it: it's a barbell. Enormous weight on one end (anything that looks like AI infrastructure, plus a handful of mega-labs), enormous weight on the other end (cheap, tiny, AI-native seed bets), and a thin, snapping bar in the middle where most companies actually live.
The bluntest version came from Max Kilberg of the seed fund Village Global, on his own show. Asked about the fundraising climate, he didn't hedge:
"Series A right now is sort of the bloodbath of the venture ecosystem for founders. It's really the haves and the have-nots. There's a lot of exciting, mostly AI infrastructure companies raising as much money as they want. Everyone else, I think, is struggling a little bit."
Village Global Podcast, "What It Actually Takes to Get a Venture Capitalist to Say Yes" with Max Kilberg (Village Global) and Michael Hochberg (Periplous) (Aug 20, 2026).
(Quick definitions, because we promised plain English: Series A is the first big "real" venture round after the initial seed money, historically the moment a startup proves it has a business, not just an idea. When Kilberg says "infrastructure," he means anything involving chips, inference, neoclouds or developer tooling, anything below the application layer. Translation: the plumbing of AI, not the apps you actually click on.)
So that's the bear case, and it has three legs.
Leg one: the money at the top is structurally hunting elephants, not rabbits. Kilberg's co-host, Michael Hochberg of Periplous, described how the biggest funds now behave: they're "only interested in deals that could absorb huge amounts of capital at multiple stages," and he called that "the dominant narrative of our time." The math is brutal and simple. When your fund is gigantic, a 100x return on a small check doesn't move the needle. You have to write enormous checks into companies that can plausibly become worth $100 billion or more. As the conversation put it, "it becomes a game of Decacorn plus hunting." (A "decacorn" is a $10 billion-plus company; they're hunting things ten times bigger than that.)
The knock-on effect for founders is nasty and rarely said out loud. When a mega-fund shows up at your seed round, they may not actually believe in you. They may just be buying a cheap lottery ticket. In Village Global's words, a seed investor now has to figure out fast "whether this company is actually breaking out or whether a multistage fund is just taking sort of a seed option check on that company because they want to be able to put in $1 billion later." You think you've been anointed; you've actually been optioned.
Leg two: the shutdown wave is already here, and it's about to get worse. On the GTM show Startup Hypeman, Erik Hammer of Marquee Ventures explained why seed has held up while everything above it wobbles, namely that you can defer the reckoning: "seed, you can kick the can. You can say, we don't really need to worry about growth rounds for five, seven years. No one knows how to price growth rounds." But all that 2021 to 2023 money eventually comes due, and Hammer has the numbers in front of him:
"I was looking at data this morning on shutdowns this quarter relative to last year. And it's about double, but I think it's going to pick up. If you look at the second half of the year, I think we're going to have a lot more companies out trying to raise when the market should tell them that they shouldn't be raising more capital."
Startup Hypeman: The GOAT to Market Show, "Best of GTM Season 3, Investing in Sports, Media, and Commerce with Marquee Ventures VC Erik Hammer" (Aug 23, 2026).
Shutdowns already running at roughly double last year's pace, and that's before the wave of overpriced growth-stage companies from the boom years hits the wall. He's candid that this is grim for the people running those companies ("it's a terrible thing to shut down a company") even as he argues it's healthy for the ecosystem.
Leg three: even great businesses are getting passed over, because of one specific fear. This was the sharpest argument of the week, from Tarek Assaad of Algebra Ventures (an Egypt-based firm) on the Making It podcast. Assaad said his fund is now walking away from companies that, by every traditional metric, look fantastic:
"We have passed on businesses that performed phenomenally well in the past three, four years, regional expansion, amazing margins, very strong moat relative to what is there today. But if we look a little bit into the future, it looks much more murky. So we've decided not to invest in these types of businesses."
The fear has a name in Silicon Valley slang: getting "Sherlocked," when the maker of the underlying platform simply absorbs your product as a feature. (The term comes from Apple copying a Mac app called Sherlock.) Assaad put the AI version of it plainly:
"What are the odds that these companies get Sherlocked? That the fundamental producer of the technology decides, 'yeah, you're a feature in my app now.' Anytime you're close to a large platform, you're running the risk of that happening."
His partner Leila had said it even more starkly back in October 2025: "We are timing out any investment that could be disrupted by AI until we have a clearer view." A whole category of perfectly good software companies is now radioactive to investors not because they're failing, but because Claude or Gemini might eat them.
Making It, "Decoding AI with Tarek Assaad" (Aug 20, 2026).
Now the other side, because the same people made it
Here's what makes this a genuine debate rather than a funeral: the bulls and bears are frequently the same investors, and their optimism is just as specific as their gloom.
The middle isn't dying, it's relocating, toward the founder. Kilberg's own counter to his bloodbath line is that seed economics still work fine: "one $10 billion outcome returns the fund and more for us, and we can also make our money on five low unicorn outcomes." And he argued the real winner of this whole reshuffle is the person building the company: "it's really never been easier to raise a $10 to $20 million seed round and really have a right to win a nascent market, to hire whoever you want, scale an engineering team, pitch the best customers." His view on defensibility is almost gleeful about the chaos: "I'm a no-moats-exist believer... ultimately the moat becomes how fast you can move. It becomes the people on your team."
This is a golden age for the lean, AI-native builder. Assaad, the same man passing on "phenomenal" companies, insists the flip side is a boom: "Across the board, this is a phenomenal time to start a company... You can do so many things with so little and do it while others haven't caught up yet." His point is that the survivors aren't the ones with big rounds; they're the ones AI can't reach. "A company that does AI for ultrasound scan will not be eaten by an LLM," because it has FDA approval, proprietary data, real-world distribution and customer intimacy. The defensible middle survives; the thin, wrapper-y middle doesn't. And crucially, he notes these companies "don't require Silicon Valley level of funding," which is exactly why the barbell's small end can still pay off.
The swing factor is a single question every founder should now ask themselves: Am I below the application layer or above it? Below it (chips, inference, neoclouds, dev tooling) and capital is nearly free. Above it, you'd better have a moat an LLM can't climb, whether regulation, proprietary data, a physical-world hook or genuine network effects, or you're a feature waiting to be absorbed. The "middle" isn't a stage of funding. It's a position in the value chain, and it's getting squeezed from both ends.
Signals
The M&A machine is running hot, and it's making paper millionaires out of two-year-old companies. The week's marquee deals: SpaceX buying the AI coding startup Cursor for roughly $60 billion, and Stripe buying the AI-model-routing company OpenRouter for roughly $7 billion (20VC, Aug 20). The 20VC crew's read on Cursor was that Elon Musk was "a shrewd buyer" who got it "dirt cheap," roughly 10x forward revenue, despite Cursor's ugly gross margins, because "your gross margin problem is my revenue opportunity for my Colossus cluster." (Cursor was selling AI tokens at a loss; SpaceX owns the data center that makes those tokens, so the loss becomes SpaceX's revenue.) The line that captures the entire risk-on mood: "Pessimists sound smart. Optimists die rich." OpenRouter is the wilder story. It raised money at a $1.3 billion valuation just four months before selling for roughly $7B, about a 5x in a third of a year, and "a 12x for Menlo and Andreessen." The bigger pattern the hosts flagged: acquirers don't buy in a panicked land-grab; they buy number two and number three because "I thought I had more time" and the leader just got taken off the table. And Stripe is doing all this, plus a rumored PayPal deal, while still private, in what the hosts called "public-company-size M&A" without the public listing.
But the sellers increasingly don't have to sell. The counterpoint came from a VC guest on RiskReversal (Aug 21): Elon reportedly tried to buy the coding startup Cognition, and they said no, because "we can raise as much money as we want in the private markets." OpenRouter, the guest argued, only got its rich price because Stripe rival Ramp had just built a competing product and forced the issue. When founders can raise unlimited private cash and sell some of their own shares (a "secondary") without giving up the company, the acquirer has to pay a premium just to get a "yes."
Anthropic turned its first profit, and 20VC explained why the accounting won't matter. Anthropic posted an operating profit on $11.5 billion of Q2 revenue, up from $4.5 billion in all of last year, with gross margins that went from negative two years ago to roughly 30% to a projected 40% (20VC, Aug 20). The hosts' point: at that growth rate, "you can't add expenses below the line fast enough to stop yourself making money." As Anthropic heads for an IPO, its filings will carry scary asterisks, including huge off-balance-sheet compute commitments and "stock-based compensation like we've never seen in the history of mankind," but the verdict was that investors will look straight past it: "none of that shit will matter. The only thing that matters will be the growth rate and the 2027 and 2028 projected revenue." The reality check was on the $200 billion and $600 billion future-revenue targets. The claim that a billion knowledge workers will each pay for AI? "Absolute bollocks." The real math: about 5 million US software workers earning roughly $600 billion combined, so hitting $200B in AI revenue means effectively replacing a third of that spend, roughly "$100,000 of tokens per engineer" while cutting dev teams 30% to 40%. Reaching $200B is imaginable. $600B "gets really hard."
Workday goes private for $43 billion, and "SaaS isn't dead" is the wrong lesson. Silver Lake circled a roughly $43 billion take-private of Workday, one of the biggest software buyouts ever, and the stock popped 18% (20VC, Aug 20). But the hosts warned against the cheerful headline. This isn't a growth story. It's "a very financially oriented, wildly savvy buyer" running a classic leveraged buyout (buy it with borrowed money, use the company's steady cash to pay down the debt): roughly 5x revenue and 16x trailing earnings, kept at 35% margins, sold in five years for a roughly 20% annual return. "So it's not dead. But what it is not is wildly exciting. This is the mature phase of an industry." A profitable retirement, not a rocket.
The coming LP liquidity flood. A theme worth watching: the RiskReversal guest argued the SpaceX, OpenAI and Anthropic IPOs will finally hand venture's backers, meaning university endowments, pension funds and hospitals, their first real cash back after years of paper markups with "no liquidity." That money then recycles into new funds: "there will probably be even more capital deployed into venture next year... where there's tons more supply, there are going to be even higher seed valuations." The paradox: more money flooding in makes it harder, not easier, to make money in seed, because "the prices are much higher and there's far more competition." It echoes the "venture has a liquidity crisis" argument Benjamin Black made on How I Invest (Aug 21), that the bar to go public has risen so far (from roughly $30M revenue in the 2000s to vastly more today) that public investors are locked out of venture's best years entirely.
The DOJ is investigating a16z, and nobody's quite sure why. On Equity (Aug 21), the TechCrunch team dug into the Bloomberg report that the Justice Department's antitrust division has been investigating Andreessen Horowitz for nearly a year. The obvious theory is board seats: a16z sits on so many boards that some portfolio companies inevitably end up competing. But two things puzzled the hosts. First, the investigation started under an administration a16z is unusually cozy with, "which makes it very surprising and strange." Second, the sheer length: "if it was just about board seats, there's not that much to investigate. There has to be some other, more serious allegation." Tellingly, a16z, normally quick to fight policy fights loudly online, has gone quiet. The read: the DOJ may be trying to "set the example" that smaller firms then follow. (Related: on 20VC, the crew noted SpaceX got fast antitrust clearance to buy Cursor as a clean corporate acquisition, partly because Meta and others now have "the DOJ crawling through what they do.")
The mid-tier is churning, and we may have hit "peak dictation app." Also from Equity: the layer of companies between the mega-labs and the scrappy seed startups is in visible turmoil. The AI automation startup Relay shut down and its team joined Google's Chrome group; Groq (the one spelled with a Q) pivoted from making inference chips to running a neocloud after an acquihire gutted its talent; and the AI dictation leader Wispr raised $280 million at a $2 billion valuation, prompting the honest question, "Have we reached peak dictation app valuations?" as the underlying tech commoditizes. Meanwhile the "AI gateway," software to manage and budget a company's AI spend, is being built by everyone at once (Databricks, Rippling, Ramp), which is exactly what drove the OpenRouter sale. The hosts' framing: the mega-labs are "plowing all this space and creating a vacuum," and the mid-tier is scrambling to find a niche before it gets sucked in.
Quote of the Week
"Series A right now is sort of the bloodbath of the venture ecosystem for founders. It's really the haves and the have-nots. There's a lot of exciting, mostly AI infrastructure companies raising as much money as they want. Everyone else, I think, is struggling a little bit."
Max Kilberg, Village Global, on the Village Global Podcast (Aug 20, 2026).
Runner-up, for the mood of the M&A market: "Pessimists sound smart. Optimists die rich." (20VC, Aug 20, 2026).